THE IMPACT OF PUNITIVE STATE IMMIGRATION POLICIES ON EMPLOYMENT AND POPULATION OUTCOMES FOR UNDOCUMENTED IMMIGRANTS A Thesis submitted to the Faculty of the Graduate School of Arts and Sciences of Georgetown University in partial fulfillment of the requirements for the degree of Master of Public Policy in Public Policy By April Carter-Chau, B.A. Washington, DC April 12, 2012
Copyright 2012 by April Carter-Chau All Rights Reserved ii
THE IMPACT OF PUNITIVE STATE IMMIGRATION POLICIES ON EMPLOYMENT AND POPULATION OUTCOMES FOR UNDOCUMENTED IMMIGRANTS April Carter-Chau, B.A. Thesis Advisor: Adam Thomas, Ph.D. ABSTRACT Illegal immigration is one of the most contentious issues of our times. The lack of comprehensive federal immigration reform has pushed states to enact provisions (e.g. Arizona s S.B. 1070 and Alabama s H.B. 56) intended to reduce their unauthorized immigrant populations. The question for policymakers is whether these bills are achieving their goals mainly to improve labor market outcomes for natives and to reduce the size of the unauthorized immigrant population. Employing data from the American Community Survey and the National Conference of State Legislatures, this analysis seeks to add to the very limited literature on the impact of recent state immigration legislation on the undocumented population. I analyze the relationship between punitive state law enforcement, omnibus, and labor immigration-related laws and the share of workers in the low-skilled sector who are undocumented immigrants. I also analyze the relationship between this legislation and the size of the undocumented immigrant population. The results show that law enforcement, omnibus, and labor legislation do not have a statistically significant relationship with the share of workers in the low-skilled sector who are undocumented. The results also show that this legislation does not have a statistically significant relationship with the size of the undocumented immigrant population. However, there is weak evidence that law enforcement legislation is associated with a small decline in the size of the population of undocumented immigrants with less than a high school education. These findings suggest that, for policymakers seeking to improve employment opportunities for natives or to reduce the size of the undocumented immigrant population, pursuing law enforcement, omnibus, and labor legislation may not be an effective approach. iii
ACKNOWLEDGEMENTS I owe immense gratitude to Professor Thomas for his encouragement, time, and love for teaching. I am grateful to Ann Morse of the National Conference of State Legislators, Madeline Zavodny of Agnes Scott College, B. Lindsay Lowell of Georgetown, Hamutal Bernstein, and Rebecca Blank, for pointing me in the right direction during my quest for data and literature and sharing their contagious passion for what they do. I also would like to thank Jeff Mayer for his miraculous way with words and Eric Gardner, Grant Blank, and Michael Barker for rescuing me repeatedly in my throes of STATA despair. I am blessed with friends, especially Heather and Brigid, who offered their editing prowess in the last week of this process and to my family who never stopped rooting for me. Most importantly, I would never have made it through these three years without the love, patience, and sacrifice of my husband. An unfathomable thank you to you all, April J. Carter-Chau iv
TABLE OF CONTENTS Introduction... 1 Background... 3 Conceptual Framework & Hypothesis... 7 Literature Review... 9 Data & Methods... 13 Data Sources... 13 Measures... 14 Descriptive Statistics... 16 Results... 23 Discussion... 38 Key Findings... 38 Policy Implications... 43 Conclusion... 44 References... 50 FIGURES and TABLES: Figure 1: Major State Immigration Legislation Categories (2006-2011)... 5 Figure 2: Simple Analysis of Immigrant Influx in a Low-skilled Labor Market... 8 Table 1: Descriptive Statistics... 18 Table 2: Examining the Relationship Between Punitive Legislation and Share of Workers in Low-skilled Sector Who Are Undocumented Immigrants... 25-26 Table 3: Examining the Relationship Between Punitive Legislation and Size of the Undocumented Population... 30-31 Table 4: Examining the Lagged Relationship Between Punitive Legislation and Undocumented Immigrants... 34-35 APPENDIXES: Appendix A. Fastest Unauthorized Immigrant Population Growth 2002-2009... 47 Appendix B. The Relationship Between Demographic Characteristics and the Size of the Undocumented Population... 48 Appendix C. Kdensity Graph for Residual Distribution Test... 49 v
INTRODUCTION More than 11 million foreign born individuals reside in the United States without documentation, 28% of the foreign-born population (Cohn & Passel, 2011). Because of the lack of success in achieving federal-level immigration reform, according to Brettel and Nibbs (2010), local governments are implementing their own measures to combat the growing presence of undocumented immigrants, who, at 8 million, constitute 5% of the American labor force. 1 Immigration experts such as Camarota (The case for, 2011) and Krikorian (Comprehensive immigration reform, 2005) argue that among the key reasons that states adopt anti-immigration legislation is to reduce the impact that illegal immigrants have on local labor market outcomes for natives and to decrease the size of the illegal immigrant population. The traditional centers of immigrant, particularly Hispanic, populations California, Texas, Florida, and New York have a long history of addressing, and, in some cases adapting to their illegal immigrant populations. However, as Anrig and Wang (2006), Massey and Capoferro (2008), and Cohn and Passel (2010) point out, undocumented immigrants migration patterns have changed over the past 2 decades. For example, according to the Census Bureau, out of the top 10 states with the highest limited English proficiency (LEP) population growth rates between 1990 and 2010, only two were also among the top 10 states with the largest LEP population. Immigrants are moving to states and urban, suburban, and rural areas that do not have a long history of immigrant populations. These states, called new mecca or new destination states, include such locations as Tennessee, North Carolina, Alabama, and Georgia. For example, Cohn and Passel, in their 2010 analysis of the unauthorized population, find that since 2000, Tennessee and Alabama have experienced undocumented immigrant growth well over 100%. According to Walters and Trevelyan (2011), immigrants who arrived after 2005 were more likely than immigrants who arrived before 2005 to live in these new mecca states. New mecca states probably are 1 For more on local government ordinances, see Brettel and Nibbs (2010); Cohn and Passel (2011) define unauthorized immigrants as foreign-born non-citizens living in the U.S. who are not legal immigrants. This group lacks the proper authorization to be legally present in the U.S. Throughout this paper I use undocumented immigrants and unauthorized immigrants interchangeably. 1
less equipped than traditional immigrant destinations to handle such demographic changes and, in many cases, have attempted to regulate these changes or prevent them from occurring by implementing restrictive immigration measures. 2 In 2011, five states passed sweeping punitive immigration legislation, parts of which the federal government is challenging in the U.S. Supreme Court (Morse, Carter, Lawrence, & Segreto, 2011). 3 Such measures, referred to as omnibus bills, are defined by the National Conference of State Legislatures (NCSL) as legislation that encompasses at least three topics that include law enforcement, education, public benefits, and employment. Illegal immigration has become highly politicized, with Congress holding numerous hearings on the impact of undocumented immigrants on natives employment and wages (The Law Library, n.d.; New jobs, 2011; Role of immigration, 2010). With or without comprehensive reform, states will be significant players because relative to the federal government, it is local governments particularly local law enforcement, according to García and Keyes (2012) that interact most often and most directly with the immigrant population. Moreover, the federal government is increasingly asking state and local governments to help implement and enforce immigration policy (National Immigration Forum, 2007). Given this expanding role for states in immigration-related issues, it is important to better understand the impacts of punitive state immigration laws. Likely because the punitive state immigration policy trend is recent, only a modest body of anecdotal and qualitative research has been conducted on the impact of state immigration legislation. Quantitative and empirical research is even more limited. This study uses extracts from the Public Use Microdata Series of the U.S. Census Bureau s American 2 For the exact immigration legislation passed by these states, see National Conference of State Legislatures (2007-2011). Restrictive immigration measures in this paper will also be referred to as punitive state policies or anti-immigration measures. Examples of these measures include fining businesses that employ undocumented workers, banning the establishment of daylaborer sites, and allowing law enforcement officers to inquire about a person s citizenship status upon an encounter with a person suspected of being undocumented. 3 Court challenges based on preemption and civil rights have been filed against the Alabama, Georgia, Indiana, South Carolina, and Utah omnibus bills, according to NCSL; for more information about these bills, see Morse, et al. (2011) State Omnibus Legislation and Legal Challenges. 2
Community Survey (2006 to 2010) to create a measure of the size of each state s undocumented population and in which occupations these populations are employed. It then identifies whether states have enacted law enforcement, omnibus, or labor legislation since 2006. By regressing the share of workers in low-skilled occupations who are undocumented and the size of states undocumented populations on this legislation, this study identifies whether there is a statistically significant relationship. Background An analysis of Congressional hearings, testimonies by immigration experts, and interviews with state representatives suggest that the salience of the illegal immigration issue reflects three key trends in America since 2006: a growing frustration with federal immigration legislation, an increasing undocumented immigrant population, and a high unemployment rate (Cohn & Passel, 2010; Immigrant gains, 2011; Morse, 2011b, 2011c, 2011d; New jobs, 2011; Role of immigration, 2010; The case for, 2011). Bruno (2010) identifies the two main sides of the current debate on U.S. immigration policy both of which intend to reduce the number of illegal aliens in the U.S. The first position seeks attrition of the undocumented population through the enforcement of immigration laws. The second approves of granting unauthorized immigrants various benefits, including a path towards legal status. Federal and State Immigration Legislation The first federal legislation to focus on illegal immigration, the Immigration Reform and Control Act of 1986 (IRCA), attempted to balance these dueling approaches. IRCA regulated the employment of unauthorized aliens by making it unlawful to knowingly hire or continue to employ unauthorized aliens without complying with the work authorization verification system (the I-9 process). The legislation also extended amnesty to almost 3 million illegal aliens. IRCA established a baseline for all future state immigration legislation. The Act focused on curtailing illegal immigration primarily by introducing employer sanctions for hiring illegal aliens. 3
Congress s modification and refinement of these sanctions in later years gradually led to the creation of a federal work-authorization system in 1996 that eventually became known as E-Verify and on which states built their own employment-related legislation. 4 States in 2005 began passing legislation intended to discourage the employment of unauthorized immigrants, for example, some legislation restricted government contracts to businesses that only employed citizens or permanent residents. IRCA s impact on the Hispanic population is not straightforward, but it is important because Hispanics make up three quarters of the undocumented immigrant population (Cohn & Passel, 2010). According to Furuseth and Smith (2006), many immigration experts point to IRCA as the critical reason behind changing demographics of Hispanic migrants in America. The authors argue that although IRCA s purpose was to decrease undocumented immigration along the U.S.-Mexico border, its amnesty provision encouraged extended family reunification and, therefore, the growth of the Hispanic population. For example, the U.S. limited English proficient population increased by 80% between 1990 and 2010, according to the U.S. Census Bureau. Parker (2010) finds that immigration is responsible for almost all of the population growth in California and 30 to 60% of the population growth in the Southwest. By increasing Hispanics movement across state boundaries, IRCA helped transform American communities and foment the development of anti-immigrant, especially anti-hispanic, environments. State officials pursuit of anti-immigration laws, therefore, in part stems on one hand from officials attempts to be responsive to constituents concerns about the changing demographics of their communities. On the other hand, Mayer (2009) argues, the pursuit reflects some state officials perception that federal immigration laws are insufficient in helping reduce the undocumented immigrant population. In 2007, NCSL reported that immigration-related laws exploded onto state legislative agendas following Congress s failure to pass any version of President Bush s comprehensive 4 For more on the history of E-Verify and employment-related legislation, see Morse, (2011a). 4
Legislation Count Figure 1. Major State Immigration Legislation Categories (2006-2011) 50 45 40 35 30 25 2006 2007 2008 2009 2010 2011 20 15 10 5 0 Education Employment ID/DL Law Enforcement Omnibus Public Benefits Legislation Category Source: NCSL, Immigrant Policy Project, 2006-2011 immigration reform. NCSL categorizes these laws into 13 categories: budget, education, employment, health, human trafficking, identification, law enforcement, legal services, omnibus, public benefits, resolutions, voting, and miscellaneous. Figure 1 shows the growth in the largest categories of state legislation with the exception of omnibus after 2006. Immigration experts agree that future Supreme Court decisions on trailblazing omnibus bills in Alabama, Arizona, Georgia, Indiana, South Carolina, and Utah will be significant to immigration s future and states ability to shape immigration policy (Morse et al., 2011). The federal government is opposing these states omnibus bills, arguing that they violate civil rights and preempt federal immigration law 5
(U.S. District Court, 2011). In addition, as of now a federal court injunction prevents the implementation of any state legislative provisions that include law enforcement s determination of an individual s lawful status, state requirements that aliens carry registration documents, state prohibitions on unauthorized aliens ability to apply for work, and state bans on transporting, harboring, or concealing illegal immigrants. Growth in the Undocumented Immigrant Population States experiencing the greatest increases in illegal immigrants in percentage terms since 2000 are Alabama (391%), Tennessee (136%), Maryland (94%), Virginia (44%), and Georgia (42%) (see Appendix A). Cohn and Passel (2010, 2011) note that the combined unauthorized immigrant population in Georgia, Arizona and North Carolina where more than one million unauthorized immigrants were estimated to reside in 2009 grew from 4% of the overall population in 1990 to 10% in 2009. The authors also find a statistically significant increase between 2007 and 2010 in the combined population of unauthorized immigrants in Louisiana, Oklahoma, and Texas. Rising Unemployment Trend Opponents of illegal immigration (i.e. supporters of the enforcement strategy) claim that undocumented immigrants reduce job opportunities for low-skilled native workers (Help save, 2012; New jobs, 2011). In addition, as Batalova, Fix, and Murray (2006) note in their analysis of immigration s impact on native workers, various polls reveal that more than one quarter of Americans believe that undocumented immigrants take native jobs. Some research supports this idea. Camarota (Immigrant Gains, 2011), for example, estimates that between 2000 and 2010, immigrant workers experienced the fastest rate of net employment growth, although natives accounted for most of the growth in the labor force. During the Great Recession of 2007, he writes, job growth for immigrants was disproportionately 6
large relative to their share of population growth when compared to job growth for natives relative to natives share of population growth. Job prospects for less-educated native workers were even worse than those for natives generally. Kochhar (New jobs, 2011) finds that native workers job loss was 3.7 percentage points larger than job loss among foreign-born workers. 5 Despite these trends, the argument that immigrants take jobs away from less-educated native workers remains controversial. Conceptual Framework and Hypothesis One of the incentives for pursuing anti-immigration policies is based on the premise that there is a relationship between the size of the undocumented population and natives job opportunities and wages. Specifically, the presence of undocumented workers reduces wages and demand for native workers. Basic labor market theory predicts that all else remaining equal, an increase in labor supply lowers the equilibrium wage (McCauley, 2008; Trebilcock & Sudak, 2006). In Figure 2, L S = labor supply, L D = labor demand, w = equilibrium wage, and Q = # of undocumented workers. An increase in the number of undocumented workers (Q* Q 2 *) shifts the labor supply curve out (L S * L S2 * ). Because there is no change in labor demand, for example, due to an economic recession, the equilibrium wage falls from w* to w 2 *. 6 Friedberg and Hunt (1995) explain that the wage decline (w* w 2 *) will cause some natives to exit the labor force or reduce their hours of work and could decrease natives employment rate. By discouraging unauthorized workers, punitive state policies attempt to mitigate this effect. Immigration experts, government officials, and employers note that restrictive measures against hiring undocumented workers raise costs for businesses that employ undocumented workers, making them less willing to hire such workers (Foley, 2010; Immigration, 2010; Lofstrom, Bohn, & Raphael, 5 From the beginning of the recession in the fourth quarter of 2007 to the fourth quarter of 2010, native-born workers lost 6.1 million jobs, a drop of 4.9%, and foreign-born workers lost 262,000 jobs, or a drop of 1.2 percent. 6 However, if all else does not remain equal, labor theory can predict a different outcome for wages. For example, if the increase in labor supply (L S * L S2 * ) due to an increase in undocumented workers (Q* Q 2 *) is accompanied by an increase in labor demand (L D * L D2 *), then equilibrium wage (w*) might not decrease. 7
2011; The business case, 2010). In other words, there is no movement along the demand curve (L D ) from the original equilibrium of (Q*, w*) to a new equilibrium of (Q 2 *, w 2 *). Likewise, Lofstrom et al. and McCauley argue that anti-immigrant law enforcement and omnibus legislation are intended to lower the unauthorized immigrant population by increasing undocumented immigrants cost of residing in a particular locality. Figure 2: Simple Analysis of Immigrant Influx in a Low-skilled Labor Market L S * L S2 * w* w 2 * L D Q* Q 2 * That is, because there is no incentive for the size of the unauthorized immigrant population to grow, there is no shift in the labor supply curve from L S * to L S2 *. This paper attempts to measure the relationship between punitive state immigration policies and labor market and demographic outcomes for undocumented workers. If state policies do not decrease the share of workers in low-skilled occupations who are undocumented or the size of the undocumented population, then these laws may not improve labor market outcomes for natives. Based on Bansak s 2005 study of the impact of IRCA s employee sanctions provision on Mexican workers (discussed in the next 8
section,) I hypothesize that state law enforcement, omnibus, and labor legislation are associated with a statistically significant decrease in the share of workers in a state s low-skilled sector who are undocumented immigrants. Based on an evaluation of the impact of 287 (g) in Prince William County, Virginia and on a study of the impact of the Legal Arizona Workers Act of 2007 (both discussed in the next section) I also hypothesize that such legislation is associated with a statistically significant decrease in the size of the undocumented immigrant population (Guterbock, Carter, Koper, Taylor, Vickerman, & Walker, 2010; Lofstrom et al., 2011). Literature Review The paucity of empirical research on the relationship between undocumented immigrants and state immigration laws reflects the newness of punitive state immigration policies and the difficulty of identifying undocumented immigrants. The limited analysis on the effect of immigration laws has typically looked either at IRCA or a single policy in a single state or jurisdiction. Studies, such as Friedberg and Hunt (1995) and Holzer (2011) that have looked more generally at immigration show that its estimated impact depends on a variety of assumptions that experts model to different degrees, such as job matching, natives in and out migration, capital movement, and factor price equalization. Because of these variations in assumptions the economic impact of low-skilled immigration is one of the most controversial topics in the immigration debate. Immigration Legislation and Undocumented Workers The literature on undocumented workers and the impact of legislation is small and what does exist has not focused on punitive policies across states. Lofstrom, Bohn, and Raphael (2011) analyze the 9
effect of the 2007 Legal Arizona Workers Act (LAWA) on unauthorized immigrants. 7 Lofstrom et al. use a difference-in-differences approach that compares the employment and population trends of Hispanic non-citizen immigrants in Arizona to the trends of Hispanic non-citizen immigrants in states that had not passed punitive-immigration legislation. To control for differences between Arizona and comparison states, the authors only use states with pre-lawa and socioeconomic trends nearly identical to Arizona s pre-lawa and socioeconomic trends. Lofstrom et al. use metropolitan statistical area fixed effects to control for other differences and distinguish between the impact of the recession and the impact of LAWA by using time fixed effects. Lofstrom et al. find that mandating E-Verify reduced the size of the proportion of Arizona s immigrant population most likely to be unauthorized low-skilled non-citizen Hispanic men but the authors do not find strong evidence that LAWA had an impact on competing authorized workers or on immigrants older than 45 years. They also find that the legislation shifted unauthorized immigrants into the informal and self-employment sectors and reduced traditional wage and salary employment opportunities for the unauthorized. Bansak (2005) attempts to identify the impact of Congress s Immigration Reform & Control Act (IRCA) on Latino workers. Bansak uses a difference-in-differences approach to compare the earnings of workers of Mexican descent relative to workers of Cuban or Puerto Rican descent before and after IRCA s passage in 1986. After controlling for Mexican ethnicity, wage, basic observable demographic and skills characteristics, and the level of employer investigations, Bansak concludes that IRCA reduced the earnings for Mexican workers the ethnic group most likely to be undocumented (Cohn & Passel, 2011) 8 relative to Cuban or Puerto Rican workers, and that this wage decline was not observed among 7 The 2007 Legal Arizona Worker Act (LAWA) mandates the use of a national identity and work authorization verification system known as E-Verify and imposes sanctions on employers hiring unauthorized workers. 8 Immigrants from Mexico have historically been the largest group of unauthorized immigrants and, as of February 2011, accounted for 58% of the total undocumented immigrant population. For more information, see Cohn and Passel, 2010. 10
non-latino White workers. The author also finds that low levels of sanctions enforcement could mitigate these effects. Orrenius and Zavodny (2009) analyze the effects of stricter national security policies implemented after 9/11 on hours worked, earnings, and employment rates of low-educated males who recently immigrated to the U.S. from Latin America a group that encompasses undocumented immigrants. The authors use a difference-in-differences approach to compare labor market outcomes for the treatment group of recently-immigrated low-educated Latin-American males relative to a control group of White non-hispanics, Hispanics, and Black non-hispanics before and after 9/11. The model controls for demographics, state-level income per capita, and employment rates and includes state and month fixed effects and an interaction term that measures the marginal effect of the business cycle on the treatment group. Orrenius and Zavodny find that most labor market outcomes among low-educated Latino males deteriorated relative to outcomes among Hispanics and Black non-hispanics. Relative to White natives, only hours worked declined for the treatment group. Most of the literature on the impact of local legislation is qualitative, localized, and anecdotal and suggest slight and inconsistent correlations between local adoption of federal government enforcement programs such as 287(g) and Secure Communities and outcomes for undocumented immigrants. 9 A 2010 Prince William County study concludes that the Hispanic, non-english speaking, and illegal immigrant population declined after the county implemented 287 (g). The study also admits, however, that it cannot discern between effects of the recession or 287(g) on the overall pattern of lower immigration (Guterbock et al., 2010). Many sheriffs whose jurisdictions have adopted 287(g) and Secure Communities across the country claim that they have seen a decrease in the foreign-born population (Green, 2010; Gwinnett, 2012; Nguyen, 2010; Whitfield, 2010). 9 The 287(g) program, administered by the Department of Homeland Security s (DHS) U.S. Immigration and Customs Enforcement (ICE), allows a state or local law enforcement entity to enter into a partnership with ICE under a joint Memorandum of Agreement (MOA) in order to receive delegated authority for immigration enforcement within their jurisdictions. Secure Communities is a federal program that partners a local jurisdiction with ICE and the FBI to share fingerprints in an attempt to identify criminal aliens. For more information visit www.ice.gov. 11
Low-skilled Immigrants and Native Workers The substantially larger body of empirical research on low-skilled immigration focuses on the relationship between low-skilled immigrants and labor outcomes for native workers. This research does not focus on undocumented workers, but because anti-immigrant legislation is predicated on the assumption that low-skilled immigrants the vast majority of undocumented workers negatively impact labor outcomes for natives, the findings of this research are important to consider. The broad conclusion of Card s work is that the impact of immigration on natives labor market outcomes is negative, but negligible (Card, 1989). In his study of the impact of the Mariel boatlift on the Miami labor market he uses a difference-in-differences approach to measure the effect of adding 56,000 less-skilled individuals to Miami s working age population and 45,000 to the labor force compared to changes in labor outcomes in comparison cities. Card does not find strong evidence of increasing unemployment rates for Whites or Blacks. This finding holds after Card controls for education, sex, marital status, part-time status, private/public employment, potential experience, and the differences between Miami and comparison cities in their employment to population rate and unemployment rate. Friedberg and Hunt (1995), in their review of research on immigrants impact on wages and growth, conclude that most of the literature finds that the effect of immigration on the labor market outcomes of natives is small. Specifically, past research also finds, that the impact of immigration on native wages clusters around zero (Borjas, 2003). Hotchkiss and Quispe-Agnoli (2008) are among the few authors that measure the impact of new waves of undocumented workers on documented and other undocumented workers. They estimate the relationship between new undocumented workers entrance into the workforce and undocumented and documented worker displacement in Georgia. After controlling for workers employed in Georgia, the number of firms, industry employment of documented and undocumented workers, the industry average 12
share of undocumented workers, annual state gross product, and industry fixed effects, the authors conclude that for all industries combined, a one percent increase in new undocumented worker hires is associated with a.03% decrease in a documented worker s likelihood of displacement and a.24% increase in an undocumented worker s likelihood of displacement. In contrast to this body of research, Borjas (2003) finds that immigration can notably reduce the labor supply and wages of competing native workers. His study introduces a new approach to estimating immigrants impact on natives labor market outcomes. Borjas uses ordinary least squares regressions with fixed effects for skill and education level and time, a two-way interaction term that controls for the variation in experience across schooling groups, and a three-way interaction term that adds state of residence to the education-experience interaction term. He finds that if one examines the relationship between immigrants and natives by different levels of education and work experience, the size of the impact of immigration on native wages can be as high as 9% for low-skilled occupations and thus suggests a larger than previously assessed reduction in the supply of native workers. Data and Methods Data Sources The unit of analysis for this study is the state-year. I created all variables except for the state-level economic and political variables by aggregating over 15 million individual-level observations up to the state level. This paper uses individual-level data from the 2006, 2007, 2008, 2009, and 2010 Public Use Microdata Series (PUMS) 1-year estimates of the U.S. Census Bureau s American Community Survey (ACS), obtained from the Integrated Public Use Microdata Series of the Minnesota Population Center 13
(IPUMS) in order to create a state-year panel dataset for the same years (Ruggles et al., 2010). 10 After aggregating, the responses were weighted using the person weights provided by IPUMS. State-level data are drawn from multiple years of the 1-year estimates of the ACS. Information about state-level immigration legislation between 2006 and 2010 is provided by the National Conference of State Legislatures (NCSL) State Immigration Legislation Database. The ACS provides employment and demographic characteristics for individuals with state-level geographic identifiers. It is the only source of national-level microdata that includes individual information, such as language ability, U.S. citizenship status, country of birth, and race/ethnicity. Because proxies for undocumented immigrants are created using a combination of most of these individual-level characteristics, such as in Lofstrom et al. (2011b) and Marra (2010), these variables can be used to more closely identify unauthorized immigrants. I focus on 16 to 60-year-olds to measure the association between state law enforcement, omnibus, and employment immigration legislation and the share of workers in the low-skilled sector who are undocumented and between this legislation and the size of the undocumented immigrant population. Measures Regression Overview. This paper uses the Ordinary Least Squares method to assess the relationship between two dependent variables and three state punitive immigration policies. The two dependent variables, the share of workers in low-skilled occupations who are undocumented immigrants and the size of the undocumented immigrant population, are estimated as follows: 10 Data was assembled using IPUMS rather than Census Bureau because IPUMS offered a simpler extraction method and the variables were easier to find. 14
Base Regression 1: EmpLow Uit = β 0 + β 1 (lawenf it ) + β 2 (omnibus it ) + β 3 (laborlaw it ) + β 4 (X it ) + β 5 (Z it ) + α i + γ t + ε it Base Regression 2: Pop U = β 0 + β 1 (lawenf it ) + β 2 (omnibus it ) + β 3 (laborlaw it ) + β 4 (X it ) + β 5 (Z it ) + α i + γ t + ε it Where i indicates the state, t indicates the year (2006, 2007, 2008, 2009, or 2010), X it is a vector of observable control variables for state-level demographic and skills characteristics that are the same across both regressions, and Z it is a vector of control variables for state-level economic and political culture characteristics that are also equivalent across both regressions. I use state fixed effects (α i) to control for all time-invariant differences between states, and year fixed effects (γ t) to control for time varying characteristics that do not differ between states. State fixed effects, for example, will control for states geographic location and political culture, two characteristics that state electoral campaigns and interviews with state legislators suggest influence the type of immigration legislation states pursue (Hakim, 2010; Morse, 2011b, 2011c, 2011d). For example, states that border Mexico may have different approaches to unauthorized immigration than states such as Alaska or Iowa have. Similarly, more conservative states, such as Alabama, may be stricter with immigration restrictions than more liberal states, such as California. Year fixed effects, for example, will control for the national unemployment rate, a statistic that might motivate state legislatures to pass employment-related legislation. ε it is the error term. Both regressions were weighted using the total population in a given state and were run using robust standard errors. 15
Descriptive Statistics Dependent Variables Table 1 contains the mean, median, standard deviation, and minimum and maximum values for the dependent and independent variables used in this analysis after applying the recommended person weight from IPUMS. The key dependent variables are the share of workers in low-skilled occupations who are undocumented immigrants and the size of the undocumented immigrant population. As mentioned earlier, it is very difficult to identify the undocumented immigrant population. Cohn and Passel (2010) employ the most widely used method for measuring the size of the U.S. undocumented immigrant population by subtracting the legal foreign-born population estimate from the total foreignborn population estimate. 11 However, there is no widely adopted method of identifying unauthorized immigrants on an individual level. Most experts use a combination of characteristics to identify a demographic group most likely to include the unauthorized population (Bansak, 2005; Borjas, 2003; Lofstrom et al., 2011; Orrenius & Zavodny, 2009). These characteristics typically include Hispanic ethnicity, less than 10 years in the U.S., working age 16 to 60, education less than or equal to high school, and male gender. Lofstrom et al. adds to these traits poor English speaking ability and non-citizenship status. I use all of these characteristics, except for gender, to create a subset of the population. If individuals meet all of these characteristics, I code them as undocumented immigrants. Although traditional measures of undocumented immigrants at the individual level limit undocumented immigrants to males, I include women because they account for 50% of my undocumented immigrant measure and because even in the aforementioned most widely used method, females in 2009 accounted for 32% of the adult undocumented immigrant population. Nevertheless, I do conduct a sensitivity analysis that tests for the exclusion of females. I discuss these results later in this paper. This 11 Foreign born population estimates are taken from the Current Population Survey and legal foreign born population estimates are taken from a variety of sources, but chiefly the Department of Homeland Security. For more information about this methodology, see Cohn and Passel (2010). 16
proxy for undocumented immigrants, after applying the appropriate frequency weights from IPUMS, results in a count of 873,566 undocumented immigrants in my dataset. To simplify the interpretation of this variable in my later regressions however, I measure this variable as the ratio of the size of the undocumented population to every 1,000 people. On average, there are 6.79 undocumented immigrants for every 1,000 people, the standard deviation is 6.03 per 1,000 people, and the maximum population for every 1,000 people is 23.5 undocumented immigrants. The first dependent variable estimates the share of the labor force accounted for by undocumented immigrants in occupations that Card (1990), Hotchkiss and Quispe-Agnoli (2008), Lofstrom et al. (2011), and Orrenius and Zavodny (2009) identify as typically attracting less-skilled workers and thus usually containing high concentrations of unauthorized workers. These occupations are manufacturing, construction, agriculture, food service, transportation, and cleaning and maintenance. I combine the different occupation responses for the categorical occupation variable in IPUMS into one variable for low-skilled occupations and estimate the number of undocumented immigrants employed in these occupations as a share of all workers employed in these occupations. 17
Variable Table 1: Descriptive Statistics Mean Std. Dev Min Max Dependent Variables Share of workers in low-skilled sector who are undocumented immigrants 0.030 0.027 0.000 0.128 Population of undocumented immigrants (per 1,000 people) 6.794 6.025 0.000 28.549 Control Variables % Married 0.548 0.050 0.270 0.635 % Minority 0.216 0.126 0.035 0.729 % Low-skilled 0.229 0.016 0.188 0.266 % Male 0.487 0.010 0.460 0.520 % HS diploma 0.505 0.048 0.390 0.632 % Non-citizen 0.066 0.045 0.006 0.207 % Foreign-born 0.116 0.078 0.017 0.356 % 10 years U.S. 0.045 0.027 0.005 0.109 % Non-English 0.147 0.101 0.022 0.459 % Hispanic 0.098 0.099 0.006 0.460 % Undoc 0.007 0.006 0.000 0.029 Total state population 5,966,414 6,679,095 515,004 37,300,000 Poverty rate 0.130 0.031 0.069 0.217 Gini coefficient 0.451 0.019 0.402 0.503 Income per capita $26,645 $3,882 $19,096 $37,871 Home vacancy rate 0.130 0.031 0.075 0.245 Employment rate 0.622 0.045 0.488 0.713 Nominal GDP (millions) $282,525 $335,813 $23,651 $1,911,741 Weeks Unemployment (average duration of unemployment insurance as % of national average) 0.960 0.117 0.703 1.315 TANF eligibility (maximum amount as % of national average) 0.991 0.431 0.318 2.211 % Democratic governor 0.547 0.499 0.000 1.000 % Republican governor 0.453 0.499 0.000 1.000 % Democratic legislature 0.535 0.148 0.190 0.903 % Republican legislature 0.461 0.150 0.088 0.810 Independent Variables Law Enforcement legislation (0,1) 0.196 0.398 0.000 1.000 Omnibus legislation (0,1) 0.033 0.178 0.000 1.000 Labor legislation (0,1) 0.282 0.451 0.000 1.000 Observations: N = 245 18
The average proportion of workers employed in low-skilled occupations who are undocumented immigrants is 3%, while the maximum share employed is 12.8%. The second dependent variable estimates the size of the undocumented immigrant population. On average, there are 6.79 undocumented immigrants per 1,000 people. 12 Independent Variables The three key independent variables in my study are Law enforcement, Omnibus, and Employment legislation. These variables are recorded in NCSL s Immigration Legislation Database. 13 Law enforcement legislation describes laws that deal with such areas as collaboration with federal law enforcement agencies, immigrant detention processes, and requiring proof of citizenship. Omnibus legislation describes a law that includes multiple measures to deter undocumented immigrants. Employment (or Labor) legislation refers to laws that discourage the employment of unauthorized workers. At 19.6%, less than half of states between 2006 and 2010 implemented law enforcement legislation. 3.3% of states implemented omnibus legislation. More than half of states 28.2% implemented some form of employment-related legislation. Control Variables The models control for a number of variables that influence in which state an undocumented immigrant chooses to live, the ability of undocumented immigrants to find employment, and the likelihood of a state passing anti-immigration ordinances. These variables are included in the vector for 12 The estimated average size of the undocumented population as I have identified undocumented immigrants in this paper is 64,675. 13 Law enforcement, Omnibus, and Labor law are binary state-level variables. Each state is assigned a 1 if the state has the legislation and 0 if the state does not. 19
state demographic and individuals skills characteristics (X it ) and the vector for state-level economic and political trends (Z it ). Demographic and skills characteristic variables are pulled from IPUMS. 14 The percentage of my sample that is male is 48.7% on average and the percentage that is married is 54.8% on average. Minorities make up an average of 21.6% of my sample, and non-citizens account for 6.6%. The average share of the population that is employed in low-skilled jobs food preparation/service, building/grounds cleaning and maintenance, personal care/services, sales, construction, maintenance/repair, and production is 22.9%. Of my sample, 11.6% is foreign-born. Half (50.6%) of my study s sample has no more than a high school diploma, 4.5% has lived in the U.S. for less than 10 years, 14.7% speaks a language other than English at home, and 9.8% is Hispanic. Undoc is the share of a state s population that consists of undocumented immigrants as I define them in this analysis. The average share is.6%, and the maximum share is nearly 3%. 15 The vector Z it controls for the impact of the 2007-2009 recession in states by controlling for home vacancy rates, state employment rates, gini coefficients, poverty rates, nominal state GDP, and state per capita income. These variables are drawn from the detailed tables of the U.S. Census Bureau s 1-year American Community Survey. The employment rates are pulled from the U.S. Bureau of Labor Statistics. Total population is a continuous variable that measures state population. I take the average of each state s population between 2006 and 2010 and use it as the analytic weight for my regression 14 These variables are weighted using the person weights provided by ACS. 15 Married, minority, low-skilled, and Hispanic originate from categorical variables that included several subgroups. The married subgroups are married spouse absent, married spouse present, separated, divorced, widowed, and single. The groups are collapsed to married, which includes married spouse absent and married spouse present, and not married. The variable is coded 1 for married and 0 for not married. The subgroups for minority are White, Black, Asian, American Indian, Asian, Mixed, and Other. These groups are collapsed to White and non-white and the variable is coded 1 for White and 0 for not White. Low-skill is created by collapsing occupations as classified according to the ACS occupational classification system into jobs traditionally considered to employ the majority of unauthorized workers (coded 1) and other jobs (coded 0). Hispanic is created from Cuban, Puerto Rican, Mexican, Other, or non-hispanic. These subgroups are collapsed to 1 if the respondent is Hispanic and 0 if the respondent is not Hispanic. Male, HS diploma, Non-citizen, Foreign-born, 10 years US, and Non-English are binary variables equal to the number of respondents in each state that possess that characteristic. This value is coded 1. 20
models. Poverty is a ratio variable that measures the proportion of a state s population that is below the poverty line. The Census Bureau uses money income thresholds that differ by family size and composition to define poverty. If a family s total income is less than the established threshold for a family of a particular size and composition, that family is considered in poverty. These thresholds are determined by the Office of Management and Budget. The average poverty rate across states for my sample is 13% (U.S. Census). The gini coefficient measures income inequality. The variable ranges from 0 to 1, where 1 is perfect inequality. In my sample the coefficient averages.45 across states. The average per capita income for my sample is $26,645 for all states in my sample. It is a ratio variable that measures the proportion of state income to state population. Vacancy rate is a ratio variable that reports the proportion of homes that are vacant. I use it to help control for the fall in housing prices that characterized the 2007-2009 recession. The average vacancy rate across states between 2006 and 2010 is 13%. The employment rate variable records the percentage of the population that is employed. Lofstrom et al. (2011) and Orrenius and Zavodny (2009), both of whom control for the impact of recessions, use this measure rather than the unemployment rate because unemployment rates are a lagged indicator for business cycles. Employment rate is therefore considered to more accurately capture the effects of an economic downturn. Nominal GDP is a continuous variable that records states nominal gross domestic production. The average nominal GDP among states in my sample is $282,525 million. Weeks unemployment insurance is a ratio variable that measures the average number of weeks that a person receives unemployment benefits as a share of the average number of weeks across 49 states. The source of these data is the U.S. Department of Labor. TANF eligibility is an interval variable that measures the maximum income allowed for eligibility for a family of three as a share of the national average maximum-allowed income eligibility. The source of these data is the Urban Institute s Welfare Rules Databook. 21
Z it also contains proxy variables that control for political attitudes toward undocumented immigrants within states. These variables are included because they control for time-varying factors that affect undocumented immigrant outcomes and because they themselves might have a significant relationship with undocumented immigrant outcomes. Leaving them out of the model would bias any correlated coefficients. These variables are governor s political party, political composition of state legislature, average duration of unemployment insurance as a share of the national average, and maximum TANF income eligibility level as a share of the national average. This model accounts for political factors because there is evidence that draws a correlation between conservative political ideals and more negative attitudes toward immigration. In their 2001 study, Hanson et al. find that political conservatives are among those most opposed to freer immigration, and that Congressional representatives seem to respond to these concerns in their districts when voting on legislation. O Neil (2010) finds that counties that voted Republican in the 2004 presidential election had a higher probability of proposing restrictive ordinances. More recently, Roy Beck, the leader of the antiimmigration group NumbersUSA predicted after the Republicans success in the November election that the anti-immigration wave would increase in potency (Carcamo, 2010). Governor s political party is a binary variable coded 0 if the party is Republican and 1 if the party is Democrat. The National Governor s Association is the source for data on governor affiliation. On average the share of Democratic governors between 2006 and 2010 was 54.7% compared to the 45.3% share of Republican governors. Legislature is an interval variable that measures the share of state legislators that is Republican and the share of legislators that is Democrat. The National Conference of State Legislatures is the source of these data. The share of states with a Democratic legislative majority on average reached 53.5% between 2006 and 2010. The share of states with a Republican legislative majority reached 46.1%. Observations for Nebraska were dropped from the model because Nebraska 22
does not hold partisan elections to its legislature. Observations for Washington, DC were also removed because the District does not have a governor or legislature. Results This section discusses my regression results, which are displayed in Tables 2 and 3. Table 2 displays the full regression results for my first hypothesis that state law enforcement, omnibus, and employment legislation are associated with a statistically significant decrease in the share of workers in low-skilled occupations who are undocumented. Table 3 displays the full regression results for my second hypothesis that this legislation is also associated with a statistically significant decrease in the size of the undocumented population. To test these hypotheses, I estimate several fixed effects multiple regression models with state-year panel data. The results of the regression analyses do not support either prediction. There is no statistically significant association between punitive law enforcement, omnibus, and employment legislation and the share of workers in the low-skilled sector who are undocumented or between these laws and the size of the undocumented population. Both tables are organized as follows: The first model (1) is the base specification. The second model (2) includes interaction terms for Hispanic, Minority, Non-citizen, and Non-English. I explain these interaction terms more fully in the Sensitivity Analysis section. The rest of the models (3-6) test the robustness of my results using different definitions of undocumented immigrants. The third model includes the same criteria as my original definition (Hispanics who do not speak English well, have no more than a H.S. education, and who are between 16 and 60- years-old) except that the gender criteria is restricted to males only. The fourth model includes the same criteria as my original definition except the language criterion is expanded to include undocumented immigrants who speak English well. The fifth model uses the same criteria as my original definition except that the education criterion is restricted such that only those with less than a H.S. diploma are designated as undocumented immigrants. The sixth 23